Pruning for Monte Carlo Distributed Reinforcement Learning in Decentralized POMDPs